search

chrisdonahue / wavegan

WaveGAN: Learn to synthesize raw audio with generative adversarial networks
MIT License
1.32k stars 282 forks source link

Cant train on 24-bit 44.1kHz wav files? #96

Open benjimyma opened 3 years ago

benjimyma commented 3 years ago

I would like to train on audio drum samples that are used in music today and the common bit depth is 24 bit and the sample rate everyone in the music industry uses is 44.1kHz. When i try training on some drum samples of these settings, i get the following exception:

ValueError: Unsupported bit depth: the wav file has 24-bit data.

I understand that 44.1kHz might be an issue to train on but this doesnt even seem to be related. seems to have an issue with the 24-bit depth of the data. Is there a fix for this or do i need to convert all my sound files to a lower bit depth ?

chrisdonahue commented 3 years ago

Try removing --data_fast_wav from your command line arguments? I believe this is a scipy issue; librosa should be able to handle 24-bit wav files